오답노트
[Keras] 연산 Layer 본문
#라이브러리 호출
import tensorflow as tf
from tensorflow import keras
import numpy as np
import pandas as pd
#데이터 불러오기
from sklearn.datasets import load_iris
iris = load_iris()
#데이터 분리
from sklearn.model_selection import train_test_split
x_train,x_val,y_train,y_val = train_test_split(x,y,test_size=0.3,random_state=2022)
x_train = pd.DataFrame(x_train,columns=iris.feature_names)
x_val = pd.DataFrame(x_val,columns=iris.feature_names)
x_train_s = x_train[['sepal length (cm)','sepal width (cm)']]
x_train_p = x_train[['petal length (cm)','petal width (cm)']]
x_val_s = x_val[['sepal length (cm)','sepal width (cm)']]
x_val_p = x_val[['petal length (cm)','petal width (cm)']]
#데이터 전처리
n_class = len(np.unique(y_train))
from tensorflow.keras.utils import to_categorical
y_train = to_categorical(y_train,n_class)
y_val = to_categorical(y_val,n_class)layers.Add
각 input을 더한다.
#레이어 선언
il_s = keras.layers.Input((2,))
h1_s = keras.layers.Dense(64,'relu')(il_s)
il_p = keras.layers.Input((2,))
h1_p = keras.layers.Dense(64,'relu')(il_p)
add_l = keras.layers.Add()([h1_s, h1_p])
ol = keras.layers.Dense(n_class,'softmax')(add_l)
#모델 선언
model = keras.models.Model([il_s,il_p],ol)
#모델 컴파일
model.compile(loss=keras.losses.categorical_crossentropy,
optimizer=keras.optimizers.Adam(),
metrics=['accuracy'])
#모델 확인
print(model.summary())
'''
Model: "model"
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_1 (InputLayer) [(None, 2)] 0 []
input_2 (InputLayer) [(None, 2)] 0 []
dense (Dense) (None, 64) 192 ['input_1[0][0]']
dense_1 (Dense) (None, 64) 192 ['input_2[0][0]']
add (Add) (None, 64) 0 ['dense[0][0]',
'dense_1[0][0]']
dense_2 (Dense) (None, 3) 195 ['add[0][0]']
==================================================================================================
Total params: 579
Trainable params: 579
Non-trainable params: 0
__________________________________________________________________________________________________
'''
그림에서 확인해보면 Add 레이어의 output의 수는 64개이다 즉 두 개의 input layer를 각각 더해서 64개가 output 된다.
layers.Concatenate
각 input을 합친다.
#레이어 선언
il_s = keras.layers.Input((2,))
h1_s = keras.layers.Dense(64,'relu')(il_s)
il_p = keras.layers.Input((2,))
h1_p = keras.layers.Dense(64,'relu')(il_p)
con_l = keras.layers.Concatenate()([h1_s, h1_p])
ol = keras.layers.Dense(n_class,'softmax')(con_l)
#모델 선언
model = keras.models.Model([il_s,il_p],ol)
#모델 컴파일
model.compile(loss=keras.losses.categorical_crossentropy,
optimizer=keras.optimizers.Adam(),
metrics=['accuracy'])
#모델 확인
model.summary()
'''
Model: "model_1"
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_3 (InputLayer) [(None, 2)] 0 []
input_4 (InputLayer) [(None, 2)] 0 []
dense_3 (Dense) (None, 64) 192 ['input_3[0][0]']
dense_4 (Dense) (None, 64) 192 ['input_4[0][0]']
concatenate (Concatenate) (None, 128) 0 ['dense_3[0][0]',
'dense_4[0][0]']
dense_5 (Dense) (None, 3) 387 ['concatenate[0][0]']
==================================================================================================
Total params: 771
Trainable params: 771
Non-trainable params: 0
__________________________________________________________________________________________________
'''
concatenate 레이어의 output을 보면 직전 layer들의 노드수를 더한 값과 같다.
Add는 노드의 값을 더했다면, Concatenate는 노드의 수를 더한다.
layers.Subtract
각 input을 뺀다.
#레이어 선언
il_s = keras.layers.Input((2,))
h1_s = keras.layers.Dense(64,'relu')(il_s)
il_p = keras.layers.Input((2,))
h1_p = keras.layers.Dense(64,'relu')(il_p)
sub_l = keras.layers.Subtract()([h1_s, h1_p])
ol = keras.layers.Dense(n_class,'softmax')(sub_l)
#모델 선언
model = keras.models.Model([il_s,il_p],ol)
#모델 컴파일
model.compile(loss=keras.losses.categorical_crossentropy,
optimizer=keras.optimizers.Adam(),
metrics=['accuracy'])
#모델 확인
model.summary()
'''
Model: "model_2"
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_5 (InputLayer) [(None, 2)] 0 []
input_6 (InputLayer) [(None, 2)] 0 []
dense_6 (Dense) (None, 64) 192 ['input_5[0][0]']
dense_7 (Dense) (None, 64) 192 ['input_6[0][0]']
subtract (Subtract) (None, 64) 0 ['dense_6[0][0]',
'dense_7[0][0]']
dense_8 (Dense) (None, 3) 195 ['subtract[0][0]']
==================================================================================================
Total params: 579
Trainable params: 579
Non-trainable params: 0
__________________________________________________________________________________________________
'''
Add 와 반대로 노드의 값을 뺀다.
layers.Average
노드 값의 평균으로 연산한다.
#레이어 선언
il_s = keras.layers.Input((2,))
h1_s = keras.layers.Dense(64,'relu')(il_s)
il_p = keras.layers.Input((2,))
h1_p = keras.layers.Dense(64,'relu')(il_p)
avg_l = keras.layers.Average()([h1_s, h1_p])
ol = keras.layers.Dense(n_class,'softmax')(avg_l)
#모델 선언
model = keras.models.Model([il_s,il_p],ol)
#모델 컴파일
model.compile(loss=keras.losses.categorical_crossentropy,
optimizer=keras.optimizers.Adam(),
metrics=['accuracy'])
#모델 확인
model.summary()
'''
Model: "model_3"
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_7 (InputLayer) [(None, 2)] 0 []
input_8 (InputLayer) [(None, 2)] 0 []
dense_9 (Dense) (None, 64) 192 ['input_7[0][0]']
dense_10 (Dense) (None, 64) 192 ['input_8[0][0]']
average (Average) (None, 64) 0 ['dense_9[0][0]',
'dense_10[0][0]']
dense_11 (Dense) (None, 3) 195 ['average[0][0]']
==================================================================================================
Total params: 579
Trainable params: 579
Non-trainable params: 0
__________________________________________________________________________________________________
'''
'Python > DL' 카테고리의 다른 글
| [DL] CNN (Convolutional Neural Network) (0) | 2022.09.19 |
|---|---|
| [Keras] 모델 성능에 유용한 Layer (0) | 2022.09.19 |
| [Keras] EarlyStopping (0) | 2022.09.15 |
| [Keras] Flatton Layer (0) | 2022.09.15 |
| [Keras] ANN (0) | 2022.09.14 |
'Python/DL' Related Articles
more
